Research On The Method Of Diagnosis And Recognition Of Pulmonary Nodules Based On Deep Learning

Lung cancer has the highest morbidity rate in the world.Pulmonary nodules,as their early manifestations,can be detected early,diagnosed early,and treated early,which can greatly reduce lung cancer mortality.With the development of medical imaging technology,the tiny lesions of the lungs in the early stage are more and more clearly displayed in front of the doctor.The continuous improvement of people's living standards and health awareness and the refined development of medical equipment have promoted the development of diagnosis and treatment of pulmonary nodules,but at the same time,it also increased the workload of traditional doctors' diagnosis.Computed Tomography examination is popular among the masses because of its noninvasive nature,but the traditional method of diagnosis and recognition of pulmonary nodules is that radiologists observe the CT slice images of the examinee 's lungs one by one.Rely on their own experience to diagnose and identify whether the patient has pulmonary nodules and to judge its benign and malignant.Facing the CT images of big data,the hospital now has the problems of time-consuming,low efficiency,and subjectivity in diagnosing pulmonary nodules.However,radiologists are prone to fatigue and prone to missed diagnosis and misdiagnosis in the face of long-term diagnosis and analysis.In view of the problems of traditional doctors in diagnosing pulmonary nodules,such as low efficiency,easy misdiagnosis and missed diagnosis,and uneven accuracy,this paper uses two deep learning methods to diagnose and identify pulmonary nodules.(1)In view of the many and complicated details of pulmonary nodules,the artificial recognition rate is not high,and it is easy to misdiagnose and missed diagnosis,a deep belief network is constructed to quantitatively study the 10-dimensional geometric data features of pulmonary nodules,they are subtlety,internal structure,calcification,sphericity,margin,lobulation,spiculation,texture,nodule volume,maximum diameter and so on.The optimal network structure and training parameters are determined through experimental research methods;The research focus on the difference between the recognition accuracy of differenthidden layers and the number of hidden nodes,and the effect of different pulmonary nodule data feature groups on the accuracy of diagnosis and recognition.(2)Make use of the spatial context feature information of pulmonary nodules,expand the pulmonary nodule samples by image preprocessing methods such as CT sequence sample stitching and enhancement operations,and construct their own sample sets;construct a diagnosis and recognition model of pulmonary nodules based on deep convolutional neural network,and discuss the basis for selecting parameters of convolutional neural network model;and determine the optimal network model structure through the study of the network model hierarchy structure method,and also discuss the effect of the number of image samples on the experiment.Through the comparison of the above two deep learning methods,the deep neural network model has excellent performance in terms of accuracy,sensitivity,and specificity.The experimental results also show that the deep convolutional neural network learning method has higher recognition accuracy.In particular,the accuracy,sensitivity and specificity of diagnosis and recognition of the constructed convolutional neural network model algorithm are 92.15%,92.17% and 93.13%,respectively.It shows that the model constructed in this paper has a good recognition effect on pulmonary nodules.In this paper,two deep learning methods based on data features and image features are used to realize the diagnosis and recognition of pulmonary nodules.The sequence images of pulmonary nodules are segmented to construct a sample set of pulmonary nodules,and then the enhanced pulmonary nodule samples are fed into the constructed deep network model,and the model is trained,tested and optimized.In this way,the automatic diagnosis and recognition of pulmonary nodule samples can be realized,which can assist doctors to make clinical diagnosis scientifically,objectively and accurately.Therefore,the research work in this paper has certain significance for the early diagnosis and recognition of lung cancer,and can provide auxiliary diagnosis conclusions for clinicians.In this paper,two deep learning methods based on data features and image features are used to realize the diagnosis and recognition of pulmonary nodules.This method is innovative.